Bio-Inspired Real-Time Robot Vision for Collision Avoidance

Abstract

A mixed analog-digital integrated vision sensor was designed to detect an approaching object in real-time. To respond selectively to approaching stimuli, the sensor employed an algorithm inspired by the visual nervous system of a locust, which can avoid collisions robustly by using visual information. An electronic circuit model was designed to mimic the architecture of the locust nervous system. Computer simulations showed that the model provided appropriate responses for collision avoidance. We implemented the model with a compact hardware system consisting of a silicon retina and field-programmable gate array (FPGA) circuits; the system was confirmed to respond selectively to approaching stimuli that constituted a collision threat.